Back to all GPUs

NVIDIA H200 Cloud GPU Specs

HexGrid Cloud rents the NVIDIA H200 SXM at $4.20 per GPU hour, billed per minute — about 8% below the $4.59/hr average across 1 other clouds. The H200 carries 141GB of HBM3e memory on the Hopper architecture.

141GB VRAMHopperDeploys in ~1 minSecure Cloud

HexGrid Cloud price

$4.20/hr

Billed per minute

VRAM

141GB

HBM3e

FP32

67 TFLOPS

Compute performance

Market average

$4.59/hr

You save 8%

Running the H200 on HexGrid Cloud

What you get when you deploy with us, beyond the hourly rate.

Live in under a minute

Pick a region, hit deploy, and SSH in. No quota requests, no sales call.

Per-minute billing

You pay for the minutes you use. Stop the instance and billing stops with it.

Ready for training day one

CUDA, PyTorch and the usual drivers are preinstalled, or bring your own image.

No egress charges

Move checkpoints and datasets out without a surprise bandwidth bill.

H200 pricing across the market

Published on-demand rates from other clouds, so you can see where our H200 price sits.

Market average $4.59/hr across 1 cloud

8% below market average
  • HexGrid CloudYou're here$4.20/hr
  • RunPod$4.59/hr

Other providers' rates are their published on-demand list prices. They exclude committed-use discounts and can change at any time. Shown for reference only.

About the NVIDIA H200 SXM

What the H200 is built for, and where it falls short.

The NVIDIA H200 extends Hopper with 141GB of HBM3e memory and 4.8 TB/s of memory bandwidth. Its larger and faster memory makes it especially suitable for LLM inference, large model fine-tuning, long-context workloads, scientific computing, and memory-intensive AI.

Best suited for

The H200 combines Hopper compute performance with 141GB of HBM3e and 4.8 TB/s memory bandwidth, making it particularly strong for large and memory-bound AI models.

  • Large LLM inference
  • LLM training
  • Long-context inference
  • Generative AI
  • HPC
  • Memory-intensive AI

Strengths

  • 141GB HBM3e memory
  • 4.8 TB/s memory bandwidth
  • FP8 Transformer Engine
  • 900 GB/s NVLink
  • MIG support

Limitations

  • High cost
  • 700W power envelope
  • Requires datacenter SXM infrastructure

NVIDIA H200 SXM specifications

Full technical specifications for the NVIDIA H200 SXM.

VendorNVIDIAArchitectureHopperChipGH100CategoryDatacenter GPUReleased2023Form factorSXMInterfaceSXM

Memory

Memory
141 GB
Type
HBM3e
Bandwidth
4,800 GB/s
Bus width
5120-bit
ECC
Yes

Compute

CUDA cores
16,896
Tensor cores
528
Tensor generation
Gen 4

AI & compute performance

FP64
34 TFLOPS
FP32
67 TFLOPS
TF32 Tensor
494.5 TFLOPS / 989 TFLOPS sparse
BF16 Tensor
989.5 TFLOPS / 1,979 TFLOPS sparse
FP16 Tensor
989.5 TFLOPS / 1,979 TFLOPS sparse
FP8 Tensor
1,979 TFLOPS / 3,958 TFLOPS sparse
INT8 Tensor
1,979 TOPS / 3,958 TOPS sparse

Power & physical

TDP
700 W
Form factor
SXM
Cooling
Server-cooled

Interconnect

Interface
SXM
NVLink
Yes
NVLink bandwidth
900 GB/s
Infinity Fabric
No

Virtualization

MIG
Yes
vGPU
Yes
SR-IOV
Yes

Media engines

NVENC engines
0
NVDEC engines
7
AV1 encode
No
AV1 decode
Yes

Software

CUDA
Yes
ROCm
No

Precision support

FP64
Yes
FP32
Yes
TF32
Yes
BF16
Yes
FP16
Yes
FP8
Yes
INT8
Yes

H200 frequently asked questions

Common questions about NVIDIA H200 SXM pricing, specifications and workloads on HexGrid Cloud.

How much memory does NVIDIA H200 have?

The NVIDIA H200 SXM has 141GB of HBM3e GPU memory.

What is H200 memory bandwidth?

The NVIDIA H200 provides 4.8 TB/s of HBM3e memory bandwidth.

Is H200 better than H100 for LLM inference?

H200 retains Hopper compute capabilities while offering significantly more and faster memory, which can improve throughput for large and memory-bound models.

Does H200 support FP8?

Yes. H200 supports Hopper FP8 Tensor Core acceleration and Transformer Engine.

How much does H200 cost per hour?

H200 hourly prices vary by provider, region, and GPU count. HexGrid compares available cloud H200 offers.

Other GPUs on HexGrid Cloud

If the H200 isn't the right fit, these are also available to deploy.

Browse every GPU we rent

Start training on the H200 today

$4.20 per GPU hour, billed per minute, running in about a minute.

Specification sources